LG Electronics is collaborating with Microsoft to create a home environment where users can interact naturally with their appliances through artificial intelligence (AI). The goal is to develop a voice agent that accurately understands user commands and responds in a human-like manner.
On September 30, Kim Cheol-ha, head of LG Electronics' HS Platform Business Center, presented the collaboration between the two companies on AI home voice agents at the Microsoft Industry Summit held in Seoul.
LG is developing AI appliances that combine on-device and cloud capabilities. In this process, the company is enhancing core functionalities of voice agents, such as speech recognition (STT), intent analysis, and text-to-speech (TTS), in partnership with Microsoft.
ThinQ ON serves as the smart home hub at the center of LG's AI home, connecting and controlling appliances and Internet of Things (IoT) devices. Based on generative AI, it engages users in everyday language, controlling connected appliances or providing necessary services based on the conversation.
Kim explained the direction of voice AI development, stating, "We are evolving to a level where the system accurately recognizes user speech and continues the conversation naturally like a human."
This approach is being applied to the voice agent of ThinQ ON. The new voice agent begins inferring and preparing responses while listening to the user, reducing conversation delays. If a user pauses, it waits for the next input, and if a new request is made during a response, it can stop the current answer and address the new request.
This real-time conversation utilizes Microsoft's Voice Live technology, which allows the system to start processing the next steps as soon as partial speech recognition results are available, thereby reducing response times.
Kim also showcased a demonstration video where the voice agent understands the context of the conversation to control appliances and generate content based on user requests. For instance, when a child asked for a comfortable sleeping temperature and requested, "Then set the bedroom air conditioner to that temperature," the system controlled the air conditioner based on the conversation.
Microsoft Azure's Custom Speech technology was employed for speech recognition, training the system to recognize LG product names and functions that are typically challenging for standard voice models.
Kim noted that using the ThinQ language pack has also reduced the time required to improve speech recognition performance across different languages. He mentioned, "In other divisions, it took about ten years to enhance speech recognition for 29 languages, but by leveraging the existing ThinQ language pack, we improved performance for 16 languages in just one year," achieving a development speed approximately 5.9 times faster than previous cases.
For speech synthesis, LG combined Microsoft's custom HD model with its AI persona to create a unique 'brand voice' that varies tone based on context and conversation.
Kim stated, "The capabilities of the voice agent are rapidly advancing." The demonstration video, filmed on September 4, initially only recommended sleep music, but the functionality has since been enhanced to allow the music to be played, along with improved response times.
He concluded, "A good voice agent understands well, converses like a person, and expresses the brand's voice. We are enriching the voice experience through our collaboration with Microsoft."
* This article has been translated by AI.
Copyright ⓒ Aju Press All rights reserved.